Public function to register custom ops (#1193)

* Added public functions to register everything.

* Removed decorator

* Revert "Removed decorator"

This reverts commit ebea5bd.

* Added some tests.

* Added the two register.

* Removed unused variables.

* Private func.

* Explicit modules.

* FLake8

* Added documentation.

* Remove useless setup method.

* Black/

* Format BUILD.

Author: gabrieldemarmiesse

tensorflow_addons/BUILD

@@ -12,6 +12,7 @@ py_library(
     data = [
         "__init__.py",
         "options.py",
+        "register.py",
         "version.py",
     ],
     deps = [
@@ -25,5 +26,18 @@ py_library(
         "//tensorflow_addons/rnn",
         "//tensorflow_addons/seq2seq",
         "//tensorflow_addons/text",
+        "//tensorflow_addons/utils",
+    ],
+)
+
+py_test(
+    name = "register_test",
+    size = "small",
+    srcs = [
+        "register_test.py",
+    ],
+    main = "register_test.py",
+    deps = [
+        ":tensorflow_addons",
     ],
 )

tensorflow_addons/__init__.py

@@ -28,5 +28,6 @@
 from tensorflow_addons import rnn
 from tensorflow_addons import seq2seq
 from tensorflow_addons import text
+from tensorflow_addons.register import register_all
 
 from tensorflow_addons.version import __version__

tensorflow_addons/register.py

@@ -0,0 +1,109 @@
+import glob
+import os
+from pathlib import Path
+
+import tensorflow as tf
+
+from tensorflow_addons.utils.resource_loader import get_project_root
+
+
+def register_all(keras_objects: bool = True, custom_kernels: bool = True) -> None:
+    """Register TensorFlow Addons' objects in TensorFlow global dictionaries.
+
+    When loading a Keras model that has a TF Addons' function, it is needed
+    for this function to be known by the Keras deserialization process.
+
+    There are two ways to do this, either do
+
+    ```python
+    tf.keras.models.load_model(
+        "my_model.tf",
+        custom_objects={"LAMB": tfa.image.optimizer.LAMB}
+    )
+    ```
+
+    or you can do:
+    ```python
+    tfa.register_all()
+    tf.tf.keras.models.load_model("my_model.tf")
+    ```
+
+    If the model contains custom ops (compiled ops) of TensorFlow Addons,
+    and the graph is loaded with `tf.saved_model.load`, then custom ops need
+    to be registered before to avoid an error of the type:
+
+    ```
+    tensorflow.python.framework.errors_impl.NotFoundError: Op type not registered
+    '...' in binary running on ... Make sure the Op and Kernel are
+    registered in the binary running in this process.
+    ```
+
+    In this case, the only way to make sure that the ops are registered is to call
+    this function:
+
+    ```python
+    tfa.register_all()
+    tf.saved_model.load("my_model.tf")
+    ```
+
+    Note that you can call this function multiple times in the same process,
+    it only has an effect the first time. Afterward, it's just a no-op.
+
+    Args:
+        keras_objects: boolean, `True` by default. If `True`, register all
+            Keras objects
+            with `tf.keras.utils.register_keras_serializable(package="Addons")`
+            If set to False, doesn't register any Keras objects
+            of Addons in TensorFlow.
+        custom_kernels: boolean, `True` by default. If `True`, loads all
+            custom kernels of TensorFlow Addons with
+            `tf.load_op_library("path/to/so/file.so")`. Loading the SO files
+            register them automatically. If `False` doesn't load and register
+            the shared objects files. Not that it might be useful to turn it off
+            if your installation of Addons doesn't work well with custom ops.
+    Returns:
+        None
+    """
+    if keras_objects:
+        register_keras_objects()
+    if custom_kernels:
+        register_custom_kernels()
+
+
+def register_keras_objects() -> None:
+    # TODO: once layer_test is replaced by a public API
+    # and we can used unregistered objects with it
+    # we can remove all decorators.
+    # And register Keras objects here.
+    pass
+
+
+def register_custom_kernels() -> None:
+    all_shared_objects = _get_all_shared_objects()
+    if not all_shared_objects:
+        raise FileNotFoundError(
+            "No shared objects files were found in the custom ops "
+            "directory in Tensorflow Addons, check your installation again,"
+            "or, if you don't need custom ops, call `tfa.register_all(custom_kernels=False)`"
+            " instead."
+        )
+    try:
+        for shared_object in all_shared_objects:
+            tf.load_op_library(shared_object)
+    except tf.errors.NotFoundError as e:
+        raise RuntimeError(
+            "One of the shared objects ({}) could not be loaded. This may be "
+            "due to a number of reasons (incompatible TensorFlow version, buiding from "
+            "source with different flags, broken install of TensorFlow Addons...). If you"
+            "wanted to register the shared objects because you needed them when loading your "
+            "model, you should fix your install of TensorFlow Addons. If you don't "
+            "use custom ops in your model, you can skip registering custom ops with "
+            "`tfa.register_all(custom_kernels=False)`".format(shared_object)
+        ) from e
+
+
+def _get_all_shared_objects():
+    custom_ops_dir = os.path.join(get_project_root(), "custom_ops")
+    all_shared_objects = glob.glob(custom_ops_dir + "/**/*.so", recursive=True)
+    all_shared_objects = [x for x in all_shared_objects if Path(x).is_file()]
+    return all_shared_objects

tensorflow_addons/register_test.py

@@ -0,0 +1,20 @@
+import unittest
+import tensorflow as tf
+from tensorflow_addons.register import register_all, _get_all_shared_objects
+
+
+class AssertRNNCellTest(unittest.TestCase):
+    def test_multiple_register(self):
+        register_all()
+        register_all()
+
+    def test_get_all_shared_objects(self):
+        all_shared_objects = _get_all_shared_objects()
+        self.assertTrue(len(all_shared_objects) >= 4)
+
+        for file in all_shared_objects:
+            tf.load_op_library(file)
+
+
+if __name__ == "__main__":
+    unittest.main()

tools/ci_build/verify/check_typing_info.py

@@ -13,12 +13,9 @@
 # limitations under the License.
 # ==============================================================================
 #
-
-from types import ModuleType
-
 from typedapi import ensure_api_is_typed
 
-import tensorflow_addons
+import tensorflow_addons as tfa
 
 TUTORIAL_URL = "https://docs.python.org/3/library/typing.html"
 HELP_MESSAGE = (
@@ -30,11 +27,18 @@
 EXCEPTION_LIST = []
 
 
-modules_list = []
-for attr_name in dir(tensorflow_addons):
-    attr = getattr(tensorflow_addons, attr_name)
-    if isinstance(attr, ModuleType) and attr is not tensorflow_addons.options:
-        modules_list.append(attr)
+modules_list = [
+    tfa,
+    tfa.activations,
+    tfa.callbacks,
+    tfa.image,
+    tfa.losses,
+    tfa.metrics,
+    tfa.optimizers,
+    tfa.rnn,
+    tfa.seq2seq,
+    tfa.text,
+]
 
 
 if __name__ == "__main__":